Listener tech spec

gistfile1.txt

  /*
   * A Listener has four states: stopped, listening, continuous (listening), and stopping.
   * (we'll handling pending/not-supported later).
   * It starts off as stopped.  It can be started normally, or in continuous mode.
   *
   * Listener Methods
   * ================
   *
   * Listener.start (normal):
   *   listener.continous = false
   *   recognizer.continuous = listener.continous
   *   recognizer.start()
   *
   * Listener.start (continuous):
   *   listener.continous = true
   *   recognizer.continuous = listener.continuous
   *   set longListenTimeout # to handle Google's API from being unresponsive after 60 seconds.
   * 
   * Listener.cancel:
   *   return if state == stopped
   *   listener.continuous = false
   *   recognizer.abort()
   *   clear abortTimeout
   *   clear finalizePendingTimeout
   *   clear longListenTimeout
   *
   * Listener.stop:
   *   return if state == stopped
   *   if state == stopping, listener.cancel() instead
   *   listener.continuous = false
   *   state = stopping
   *   recognizer.stop()
   *   if pendingResult:
   *     Should finalize pendingResult and emit? Would that duplicate?
   *   set abortTimeout # to handle chrome bug where it doesn't always end
   *   clear longListenTimeout
   *   clear finalizePendingTimeout
   *
   * 
   * Recognizer events
   * =================
   * onaudiostart: ignored
   * onsoundstart: ignored
   * onspeechstart: ignored
   * onspeechend: ignored
   * onsoundend: ignored
   * onaudioend: ignored
   * onnomatch: ignored for now
   *
   * onresult:
   *   clear finalizePendingTimeout
   *   if result.final:
   *     pendingResult = null
   *     emit result, naturalFinal = true
   *   else:
   *     if listener.interimResults:
   *       emit result
   *     pendingResult = result
   *     set finalizePendingTimeout
   *
   * onerror:
   *   "no-speech": ignore
   *   "aborted": ignore
   *   "audio-capture": emit
   *   "network": emit
   *   "not-allowed": emit
   *   "service-not-allowed": emit
   *   "bad-grammar": emit
   *   "language-not-supported": emit
   *
   * onstart:
   *   state = listening
   *   emit start
   *
   * onend:
   *   if continous == true:
   *     recognizer.start()
   *   else:
   *     state = stopped
   *     clearAbortTimeout
   *     emit end
   *
   */

#9: we'll also set the finalresult time out here

#34: it should emit IMO bc the result was never finalized/emitted.

#63: we fire this event currently right? It's annoying though bc we force abort a lot

Can you also add descriptions/functionality for the 3 timeouts we have: longListen, onEndAbort, and setEarlyFinalResult?

Thanks for putting this together. I think this gist should be our source of truth for what the listener should behave.

#59: add if continuous

#34: move to recognizer.onend